08:02 asyync1024: Hey! I have returned after installed nouveau on my GT 740, Kepler, NVE0, Wayland. This time, turning on the 0f performance state for the first time caused a logout, but when I tried again with a reboot, it sticked but things were slightly buggy, I fired up firefox/other programs and it gave another logout.
08:05 asyync1024: Means the reclocking support definitely has made some progress, rather than dying instantly, it just logs out and keeps the performance state, though opening apps like firefox causes a crash. I have managed to get a log through "journalctl -k -f -o short-monotonic > /var/log/nouveau-kernel-live.log" run with systemd-run.
08:06 asyync1024: Here is the pastebin of the output I got, after firefox crash. I took this output after a reboot: https://pastebin.com/JJZ3GLHd
10:03 x512[m]: About Vulkan UVM: https://github.com/KhronosGroup/Vulkan-Docs/issues/2676
10:04 x512[m]: Maybe it is even possible to introduce Vulkan UVM extension that will replace existing memory management API, similar to descriptor heap extension.
13:33 phomes_[d]: I am looking into a very slow shader
13:33 phomes_[d]: It has a const array of 733 vec3's. NVK moves these to st.local causing 2k mov's. Prop just does ldc c[0x0][]
13:34 phomes_[d]: I wonder where a fix for that should go
13:47 karolherbst[d]: uhhhhhhhh
13:47 karolherbst[d]: yeah sooo
13:47 karolherbst[d]: the issue is we can't do indirects on const arrays
13:47 karolherbst[d]: `nir_opt_large_cosntants` I think...
13:48 karolherbst[d]: but...
13:48 karolherbst[d]: weird that it doesn't promote it to a cbuf
13:49 karolherbst[d]: phomes_[d]: yeah.. please file an issue and attach a `NIR_DEBUG=print` for the shader especially around nir_opt_large_constants
13:55 karolherbst[d]: actually.. I was running into similar issues 🙃
13:55 karolherbst[d]: but I don't know if the constants were huge enough...
13:55 karolherbst[d]: anyway, good find
14:00 phomes_[d]: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14993
14:03 karolherbst[d]: phomes_[d]: oh I meant `NIR_DEBUG=print`
14:03 karolherbst[d]: not NAK_DEBUG
14:03 phomes_[d]: my bad. 2 sec
14:16 phomes_[d]: I have added it to the issue now. No output at nir_opt_large_constants though
14:20 karolherbst[d]: yeah...
14:20 karolherbst[d]: I have a theory already (without having looked at it)
14:20 karolherbst[d]: just wanted to confirm
14:26 karolherbst[d]: ahhh yeah..
14:26 karolherbst[d]: pain 😄
14:27 karolherbst[d]: `nir_lower_variable_initializers` 🥲
14:27 karolherbst[d]: yeah sooo
14:27 karolherbst[d]: uhm...
14:27 karolherbst[d]: yeah...
14:32 karolherbst[d]: phomes_[d]: I've attached a patch that might help?
14:43 phomes_[d]: it did. Compiles to much smaller size and game perf is great
14:43 phomes_[d]: but it does cause other problems like black screen flicking
14:44 karolherbst[d]: mhhh
14:45 karolherbst[d]: yeah maybe I look into it later or somebody else does, not sure I really have time for it anyway